Tech credit spreads have flipped from the tightest in the corporate market to wider than average, as hyperscaler bond supply passes $220 billion for the year
NEW YORK, August 21: The debt financing artificial intelligence infrastructure is meeting resistance from the investors who have absorbed it all year. Fund managers are still comfortable with the credit quality of the companies borrowing, but they are asking to be paid more for taking down deal after deal from the same handful of names.
Tech corporate bond spreads sit at 89 basis points, roughly 9 basis points wider than the broader investment-grade market, Capital Group portfolio manager Karen Choi told Reuters. That is a reversal for a sector that historically borrowed at some of the tightest levels in credit, backed by heavy cash balances and modest funding needs.
Spreads measure the extra yield investors demand over US Treasuries. Wider means more perceived risk or, in this case, more compensation for absorbing supply.
A Sector That Now Borrows Like a Bank
The composition of the investment-grade index has shifted fast. Banking remains the largest sector at around 21% of the Bloomberg US Corporate Index, while technology has climbed to roughly 10% on the back of AI-driven supply, and in several major indices technology borrowers have overtaken banks in weight for the first time.
The supply figures explain why. Between 2020 and 2024, the five largest hyperscalers issued roughly $35 billion of debt each year on average. In 2025 that figure jumped to $93 billion. Through July 31 this year they had issued approximately $132 billion, including one multitranche offering of roughly $53 billion, among the largest corporate bond sales on record, and a rare century bond maturing 100 years from issuance.
Tallies vary depending on what gets counted. BNP Paribas data cited by Reuters put hyperscaler issuance at $220 billion for 2026 as of August 10. A Reuters analysis of LSEG data found Amazon, Alphabet, Meta and Oracle sold about $194 billion of bonds through early July, up 79% from roughly $108 billion across all of 2025. Goldman Sachs strategist Amanda Lynam estimated global AI-related debt issuance at $489 billion this year, already above the bank's $322 billion estimate for all of last year, with about 40% of it issued directly by hyperscalers.
Order Books Are Thinning
The clearest evidence of fatigue is in how deals clear.
The coverage ratio for hyperscaler bond issuance fell from 5x in February to below 2x by July. Amazon's eight-part offering in July is the reference point analysts keep returning to. Demand for the $25 billion deal peaked at $62 billion, then was pared back to roughly $41 billion once banks tightened the final spread, leaving the book at about 1.6 times deal size. Pricing ran from 40 basis points over Treasuries on the three-year tranche to 125 basis points on the 40-year, with the 30-year at 110.
Reuters reported that analysts pointed to Amazon's long-dated sale clearing at roughly 120 basis points over Treasuries, about double where comparable paper would have priced a year earlier.
The knock-on effects were immediate. The offering came as a surprise to the market and pushed Amazon's existing 30-year bond, issued earlier in the year, 20 basis points wider as investors demanded additional concession. Hyperscaler bond spreads widened between six and 15 basis points that day, and the deal helped lift the 10-year Treasury yield by eight basis points, according to Bank of America.
Alphabet's turn came a month later. The company sold $25 billion of investment-grade bonds on August 6, drawing roughly $115 billion of peak demand, behind only the record $129 billion for Oracle's February deal and the roughly $126 billion Amazon's March sale attracted. The offering spanned 10 tranches with maturities of two to 40 years, with initial talk of about 155 basis points over Treasuries on the 40-year piece. Reuters reported that the deal was well received but still required a concession of about 10 to 15 basis points against Alphabet's existing bonds.
New-issue concessions in the sector have been running far above the market norm. They have averaged around 12 basis points for AI-related deals, against roughly 2.5 basis points for the broader market.
The Cost Shows Up Offshore Too
Hyperscalers have spread their borrowing across currencies to find fresh buyers, and the strain is visible there as well.
Alphabet paid just under 7% on the longest portion of its debut Australian dollar bond on August 19, its highest yield ever on a note. The A$5.5 billion deal, worth about $3.9 billion, priced its 20-year tranche at 6.98%, while almost two thirds of the offering came in maturities of five years or less at fixed rates of roughly 5.2% to 5.5%. Bank of America found the non-dollar share of hyperscaler issuance doubled to 30% by 2026, with record-setting deals in euros, sterling, yen, Swiss francs and Canadian dollars.
The 2% Problem
The harder constraint may have nothing to do with credit quality.
Choi told Reuters that many pension and insurance investors cap exposure to any single issuer at roughly 2% to 3% of assets. When the same small group of borrowers returns to the market repeatedly, those internal limits start to bind regardless of how the fundamentals look. She noted that clients do not want to open a statement and find a single bond dominating their holdings.
For now, foreign buyers, pension funds and insurers have absorbed much of the paper, helped by an investment-grade index yielding around 5.4%, in line with long-run averages, according to Reuters.
JPMorgan strategists have framed the recent widening as high-grade investors rationally repricing an accelerating pace of supply rather than a verdict on the borrowers. Apollo chief economist Torsten Slok has raised a blunter question: who becomes the marginal buyer of investment-grade paper if hyperscaler supply keeps climbing.
Fundamentals Are Strong, Cash Flow Less So
Ratings and balance sheets remain the bull case. Post-issuance debt levels for hyperscalers often sit in the 0.4 to 0.7 times range against an investment-grade average near 3 times. Only Oracle and SpaceX carry BBB ratings within the cohort, with the rest in the single-A to double-A range.
Cash generation is a different picture. Aggregate free cash flow across the five largest names fell around 24% in 2025 compared with 2024, with further pressure expected this year. Amazon's trailing twelve-month free cash flow dropped to $1.2 billion from $25.9 billion a year earlier, driven by a $59.3 billion year-on-year increase in property and equipment purchases, though operating cash flow still came in at $148.5 billion, up 30%. Oracle's five-year credit default swap has become the market's preferred proxy for AI debt anxiety and has been trading at a multi-year high.
Capex Guidance Points to More Supply
Nothing in company guidance suggests the issuance calendar thins out.
UBS lifted its 2026 US investment-grade issuance forecast to $1.8 trillion and raised its technology supply estimate to $360 billion from $300 billion, after Amazon, Meta and Google each raised capital expenditure guidance materially above consensus. That took projected hyperscaler capex to around $770 billion, about 23% higher than the bank previously expected, implying an additional $40 billion to $50 billion of public debt issuance. Alphabet alone raised more than $50 billion in debt during the first half of 2026 and issued nearly $85 billion of equity earlier in the year.
"It's not a blank check," DWS head of fixed income for the Americas George Catrambone told Reuters, warning that concessions and spreads both widen if companies keep returning to the same well.
Goldman Sachs expects bond issuance from the five largest hyperscalers to reach $250 billion this year and $400 billion in 2027, with AI-related debt already accounting for roughly 15% of total investment-grade issuance, according to Barclays data. Aggregate hyperscaler capital expenditure is projected to approach $800 billion this year and exceed $1 trillion annually from 2027 through 2030.
Comments
Join the discussion and share your perspective.